Developers and enterprises training large language models (LLMs) and deploying AI workloads in the cloud have long faced a fundamental challenge: it’s nearly impossible to know in advance if a cloud platform will deliver the performance, reliability, and cost efficiency their applications require. In this context, the difference between theoretical peak performance and actual, real-world results is often the difference between progress and frustration.
A lack of transparent benchmarking practices, inconsistent results and performance across cloud providers, and no clear standards often leave teams making critical infrastructure decisions in the dark. What does “good” performance really mean? How can you compare across clouds? How do you know you’re getting what you paid for? Should reliability be a factor?
Today, NVIDIA is addressing these challenges with the launch of NVIDIA Exemplar Clouds. This new initiative is designed to bring transparency, rigor, and reproducibility to the world of AI cloud infrastructure, starting with the NVIDIA Cloud Partner (NCP) ecosystem.
NCPs are specialized cloud providers who build their platforms around the latest NVIDIA GPU architecture, software stacks, and best practices. Until now, there’s been no standardized or public way to verify that these platforms are tuned for real-world AI workloads or exhibit resiliency needed for developer productivity. NVIDIA Exemplar Clouds addresses this by introducing frameworks that evaluate providers on actual performance and resiliency, not just theoretical specs.
Benefits of NVIDIA Exemplar Clouds
With NVIDIA Exemplar Clouds, every participating cloud provider undergoes a comprehensive evaluation process designed to reflect real-world customer needs and operational excellence. Achieving Exemplar status requires NCPs to demonstrate high performance and resiliency across a suite of open, workload-specific benchmarking recipes covering workloads from inference, fine-tuning, and scaled pretraining. The result is a transparent, apples-to-apples comparison that empowers customers to make informed decisions based on performance and TCO.
NVIDIA also shares benchmarking recipes and results through NVIDIA DGX Cloud Benchmarking, part of the criteria for the NVIDIA Exemplar Clouds initiative. Providing this workload-by-workload transparency empowers developers, researchers, and enterprises to hold providers accountable and to optimize their own deployments with confidence. This approach also enables NCPs to continuously improve their platforms based on clear, actionable feedback.
Figure 1 compares total cost and total time to train a model using FP8 versus BF16, as seen in NVIDIA DGX Cloud Benchmarking Performance Explorer. For more details, see Measure and Improve AI Workload Performance with NVIDIA DGX Cloud Benchmarking.
Apart from creating the evaluation framework, NVIDIA will also work with NCPs to earn Exemplar status, bringing its suite of software, tools, and processes. The framework will evaluate true workload performance, resiliency, user access, security, and more.
Spotlight: Yotta
NVIDIA is excited to announce that Yotta is the first APAC cloud provider to join the NVIDIA Exemplar Clouds initiative. As India’s AI cloud provider, Yotta has demonstrated the ability to deliver consistent, high-performance results across a range of demanding AI workloads.
Over the course of the next few months, the NVIDIA team will work with Yotta on aspects of user experience, performance and resiliency. With the help of the NVIDIA Exemplar Clouds initiative, Yotta’s customers can now access detailed benchmarking data, see exactly how their infrastructure performs for each use case, and have confidence that their workloads will run as expected—no guesswork, no surprises.
The NVIDIA team plans to work closely with Yotta in the coming months to benchmark workload performance, resiliency, security, and other aspects that drive AI user experience.
Get started
NVIDIA Exemplar Clouds democratize access to world-class AI infrastructure. Sovereign AI initiatives require a new standard for transparency and performance, and we’re helping the entire industry move forward, faster. Exemplar Clouds help every user—regardless of geography or size—can build and deploy AI with confidence. If you’re building mission-critical AI and want to optimize your AI workloads, get started by exploring NVIDIA DGX Cloud Benchmarking.
To learn more, join NVIDIA founder and CEO Jensen Huang for the COMPUTEX 2025 keynote and attend GTC Taipei sessions at COMPUTEX 2025 through May 23.
